|
Description:
|
The problem of accurately locating the position of an animal using noisy directional data is treated here from the viewpoint of extended Kalman filtering methodology . The physical model considered consists of an animal moving randomly in a confined area and the location of it is tracked using several fixed measuring devices , each of which is nominally capable measuring the angular location of the animal from its own location . These angular measurements are inaccurate due to random noise . The system is modelled mathematically as follows . Time is assumed to move in discrete steps which coincide with moments at which measurements are taken . During a given time step , the movement of the animal is described by a linear difference equation driven by random noise , and the inaccuracies of the measuring devices are modelled as additive noise with zero mean and known covariance . The extended Kalman filtering problem for this system is formulated , and theoretical analysis is carried out . It is shown that if the animal movement is suitably confined , then the covariance of estimation errors satisfy a stable dynamical system . In particular , bounds on the magnitude of the covariance of estimation errors are derived . It is also shown that there is an associated extended Kalman filtering problem with stable filter and covariance dynamics . Extensive simulation experiments are carried out to compare the performance of Kalman filtering strategies with well established triangulation methods . |